AI-assisted marking in higher education

Universities are rapidly exploring generative AI for assessment, yet student acceptance hinges on perceptions of fairness, transparency, and human oversight. This study examines how students evaluate AI-assisted marking configurations in higher education. We fielded a scenario-based, choice-based conjoint experiment with 340 students, who repeatedly chose between assessment options varying on five attributes: marker type (human, hybrid, AI-only), recognition of creativity, objective reliability evidence, feedback aggregation, and turnaround time. Hierarchical Bayesian estimation revealed marker type as the dominant driver of choice: hybrid human–AI marking was consistently preferred, whereas AI-only options were strongly rejected. Recognition of creativity and objective reliability both exerted large, independent positive effects. Aggregated feedback and quicker turnaround improved acceptance, but their effects were smaller and contingent. Exploratory latent-class analysis and moderation by enrolment motivation indicate meaningful heterogeneity: for students seeking novel learning experiences, creativity recognition was particularly salient; for compliance-oriented or self-development motivations, faster, aggregated feedback mattered more. We conclude that acceptable AI in assessment is designed, not assumed: human oversight should remain visible and consequential; systems should make reliability auditable; and feedback should be transparent about sources and aggregation. We discuss implications for assessment governance, including disclosure to students, moderation and appeals, and phased implementation of hybrid models that foreground human accountability while leveraging AI for scale and timeliness. Findings extend acceptance research by isolating design levers generalising across grading contexts and diverse student segments in universities.